Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77300.8975367836,
-      "required_pages_per_second_upper_estimate": 0.005347855454388345
+      "required_token_positions_per_second": 77466.24365159356,
+      "required_pages_per_second_upper_estimate": 0.005359294482266742
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9662.61219209795,
-      "required_pages_per_second_upper_estimate": 0.004045533946543866
+      "required_token_positions_per_second": 9683.280456449194,
+      "required_pages_per_second_upper_estimate": 0.004054187317225294
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19325.2243841959,
-      "required_pages_per_second_upper_estimate": 0.00402493459878916
+      "required_token_positions_per_second": 19366.56091289839,
+      "required_pages_per_second_upper_estimate": 0.004033543907600319
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28987.83657629385,
-      "required_pages_per_second_upper_estimate": 0.0017377765408852091
+      "required_token_positions_per_second": 29049.841369347585,
+      "required_pages_per_second_upper_estimate": 0.0017414936335529436
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28987.83657629385,
-      "required_pages_per_second_upper_estimate": 0.002253676789788788
+      "required_token_positions_per_second": 29049.841369347585,
+      "required_pages_per_second_upper_estimate": 0.002258497389718454
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28987.83657629385,
-      "required_pages_per_second_upper_estimate": 0.3718248428867491
+      "required_token_positions_per_second": 29049.841369347585,
+      "required_pages_per_second_upper_estimate": 0.37262017379648266
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10097817867968928,
-    "producer_p95_seconds": 0.20196183299958648,
-    "gpu_update_seconds": 0.9537793524960892,
-    "producer_batches_per_second": 9.903129696685049,
-    "gpu_batches_per_second": 1.0484605243161838,
-    "producer_headroom": 9.44540062978906,
+    "producer_batch_seconds": 0.282047311867224,
+    "producer_p95_seconds": 0.19482041700030095,
+    "gpu_update_seconds": 0.9517435791981033,
+    "producer_batches_per_second": 3.5455044523549932,
+    "gpu_batches_per_second": 1.0507031745279074,
+    "producer_headroom": 3.3744110975471524,
     "sustainable": true,
-    "buffer_seconds": 60.08809920725362,
+    "buffer_seconds": 60.91158906867861,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372771.36241697177,
-    "upload_bytes_per_second": 4463851.165853077,
+    "generation_bytes_per_second": 368136.1855508523,
+    "upload_bytes_per_second": 4545068.308342919,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.98805670148721,
+    "estimated_full_shards_per_hour": 78.99345564741303,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -532,6 +532,12 @@
       "bytes": 856828,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.4432682090000526,
+      "bytes": 13357057,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -70,3 +70,4 @@
 {"step": 849500, "loss": 0.5055427551269531, "lr": 3e-05, "grad_norm": 0.29385733964605903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124343326771, "tokens_per_second": 152961.74688759536, "data_wait_s": 1.499599966336973e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849600, "loss": 0.48633232712745667, "lr": 3e-05, "grad_norm": 0.2856631745968521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124358058707, "tokens_per_second": 153195.3083980939, "data_wait_s": 1.4041000213182997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849700, "loss": 0.49540627002716064, "lr": 3e-05, "grad_norm": 0.2918645114072405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124372663625, "tokens_per_second": 152429.32540568532, "data_wait_s": 1.3471000784193166e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 849800, "loss": 0.5476540923118591, "lr": 3e-05, "grad_norm": 0.2951122253907351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124387247556, "tokens_per_second": 151878.84600292164, "data_wait_s": 1.462500040361192e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b94dabca98c78f943feee45e69a84fca13b1847424396a56237bb7ac8cf6711b
-size 153444
+oid sha256:bd2b328a938f826870b5f2c794cdbe2a706574150b51e07f8b6e3855574ff4fa
+size 155562